August 25, 2026, (Inside AI) — A pair of AI researchers who walked away from a Bezos-backed venture have unveiled a physics-first model that processes 5 trillion data points in one prompt. The company, Accelerated Understanding Inc, says its system is not built for language but for predicting physical phenomena.
The co-founders, Anima Anandkumar and Benedikt Jenik, revealed the technology in an exclusive interview ahead of Tuesday’s corporate launch. Anandkumar, a Caltech professor, helped pioneer neural operators years ago, a method that replaces the Transformer architecture used in ChatGPT.
Their model’s capacity dwarfs flagship systems from Anthropic and Google. It can ingest roughly 5 million times more data in a single prompt, equivalent to reading ‘War and Peace’ 5 million times in one sitting. The difference lies in its focus: physics, not text.
“The language-centric view of intelligence is humans at the center. Putting physics at the center is a nature-centric view,” Anima Anandkumar, co-founder, Accelerated Understanding Inc
Read: China’s Fysics AI Launches Physics-Based World Model, Bypassing US AI Road Maps
This approach aligns with a broader push toward world models, pursued by startups linked to AI leaders Yann LeCun and Fei-Fei Li. Accelerated Understanding bets that a more general neural operator can predict unseen phenomena and unlock enterprise value.
Early applications include chip design, where intrinsic physics grasp could optimize materials and temperatures with less lab trial and error. The same AI could power robotics, predict extreme weather, or sift geological data for energy firms. The company targets enterprise deals first, not consumer products.
Bezos Dinner Led to a Fork in the Road
In late 2024, Anandkumar and Jenik dined with investor Vik Bajaj at an upscale Los Angeles restaurant. Bajaj would later co-found Project Prometheus with Jeff Bezos. Meeting records show they discussed a collaboration.
An offer letter titled ‘Project Prometheus’ proposed Anandkumar as public face, board member, and owner of scientific vision. Jenik would be a board observer. The deal included a 35% stake and a combined $1 million annual salary, doubling after three months.
The letter outlined more than $2 billion in committed financing through Series B, with Bezos among investors. Anandkumar and Jenik declined. Bezos and Bajaj went on to raise a $12 billion Series B for Prometheus in June 2026, targeting AI for manufacturing complex physical systems.
Prometheus declined to comment on the offer.
Read: VeloAlpha’s FusionAlpha Uses AI to Crack Fusion Energy’s Software Bottleneck
Nvidia Roots and a Bold Prediction
Anandkumar’s path began at Nvidia, where she was hired in 2018 and led a team advancing GPUs for frontier AI. An early project showed AI could speed weather prediction with accuracy matching complex forecast computations. Jensen Huang presented her neural operator work at GTC 2021.
“He just got so excited,” Anima Anandkumar, co-founder, Accelerated Understanding Inc
When she said AI could eat physics theorists’ lunch, Huang replied: “I want it to eat all their lunches.” Anandkumar declined to discuss funding but confirmed partnerships with computing providers for hardware clusters. She did not name them. Nvidia did not respond to questions about backing the endeavor.
Accelerated Understanding’s bet is that physics-centric AI can outperform text-trained models on real-world tasks. The company’s launch signals a shift from language models to systems that understand space and time. Whether enterprises adopt this approach remains an open question, but the founders’ pedigree and early traction suggest a serious contender in the race for world models.
